intelligent agents -...

intelligent agents

Eleni Kaldoudi

Researcher, ICS-FORTH

C-472: Machine Vision

Department of Computer Science

University of Crete at Heraklion

13 February 1997

Eleni Kaldoudi, ICS-FORTH

intelligence

R.A. Brooks (MIT)

“… the sort of stuff humans do,

pretty much all the time” (!)

intelligence is determined by the

dynamics of interaction with the world

intelligence is in the eye of the observer


traditional approach

top-down approach

thought and reasoning - introspection

– I/O to thoughts and reasoning (knowledge representation)

– what thoughts and reasoning do?

(planning, problem solving,…)

static environment

closed system

sense-model-plan-act

define and control a search

space


key realizations (around 1984)

situatedness

embodiment

intelligence

emergence

dynamic environment

interaction with the

environment

collaboration of system

components

agents * active vision * animate vision

artificial creatures * behavior-based approach


what is an agent ?

self-contained program

capable of controlling its own decision making and acting,

based on perception of complex and dynamic environments

and in pursuit of one or more goals

a system that can be most conveniently described by the

intentional stance, by ascribing mental qualities

information attitudes

belief

knowledge

pro-attitudes

desire

intention

obligation

commitment

….


agents’ characteristics

agent: goals, sensors, effectors, environment

autonomy

social ability

reactivity

pro-activeness

adaptivity

interaction dynamics

environment

agent components

agent society

emergent intelligence


characteristics of agent architectures

task-oriented modules

task-specific solutions

de-emphasized representations

de-centralized control

emergent goals and activity

learning - redundancy


two interesting problems

action - selection

learning from experience


action selection: the problem

assume: multiple goals,

specific sensor data and

a variety of possible actions

what are the next actions in order to

optimize the achievement of goals ???

favor maximum progress

minimize contradictory

actions

take opportunities

be real-time

improve by experience

when things go wrong,

degrade, don’t break down

avoid endless loops

be adequate

requirements


action selection: towards a solution

a theoretically computed action selection policy

is not possible

• resource limitations

• incomplete and incorrect information

• dynamic, non-deterministic environment

• time-varying goals

• ….

solutions depend on • nature of goals

• nature of sensor data

• command control and fusion


learning: the problem

assume: multiple, time-varying goals,

specific sensor data and

set of actions or competence modules

how can the agent improve the action selection

behavior based on experience ???

incremental learning

selective learning

unsupervized learning

cope with noise, faulty sensor data, etc.

possibility for built-in knowledge

requirements


learning: towards a solution

focus of attention

perceptual selectivity

what stimuli might possibly be attended to any time

cognitive selectivity

what internal structures are attented to at any time


applications

Interface agents

Information Agents and Cooperative Information Systems

Believable Agents

Cooperative Problem Solving and Distributed AI

power systems management

air-traffic control

particle accelerator control

intelligent document retrieval

patient care

telecommunications network management

spacecraft control

computer integrated manufacturing

transportation management

job shop scheduling

steel coil processing control


more details in ...

R.A. Brooks, “Intelligence without Reason”, A.I. Memo no. 1293, AI Lab, MIT,

1991 (http:// www.ai.mit.edu)

P. Maes, “Modeling Adaptive Autonomous Agents”, Artificial Life Journal,

ed. C. Langton, vol. (1&2), MIT Press, 1994.

(http:// www.media.mit.edu)

N.R. Jennings, M. Wooldridge, “Software Agents”, IEE Review, Jan. 96, 17-

20, 1996. (http://www.elec.qmw.ac.uk/dai)

M. Wooldridge, N.R. Jennings, “Intelligent Agents: Theory and Practice”,

Knowledge Engineering Review, vol 10(2), 115-152, 1995.

(http://www.elec.qmw.ac.uk/dai)

Y. Shoham, “Agent-Oriented Programming”, Artificial Intelligence, vol 60,

51-92, 1993.

L.N. Foner, “Paying Attention to What’s Important: Using Focus of Attention

to Improve Unsupervised Learning”, MSc Thesis, MIT, 1996

http://ksl/stanford.edu


cite as

E. Καλδούδη, “Πράκτορες Λογισμικού (Software Agents) –

Εισαγωγή”, Σεμινάριο στα πλαίσια του μαθήματος “C-472:

Μηχανική Όραση”, Υπεύθυνος: καθ. Σ. Ορφανουδάκης, Τμήμα

Επιστήμης Υπολογιστών του Πανεπιστημίου Κρήτης, 13

Φεβρουαρίου 1997

intelligent agents -...

Documents